FunGT Mesher is a C++ and SYCL application for generating 3D data from a single image. It implements the inference pipeline on the GPU using pretrained TripoSR model weights, encoding the input image, generating triplane features, and producing a density grid that can be converted into a 3D mesh.
Set the model weights path in the environment:
export MODEL_PATH=/path/to/model_weightsThe committed fungt-mesher.config reads this variable:
model_path=${MODEL_PATH}Run the generator with an input image and output density file:
./FunGTMesher image.png --out image.objFunGT Mesher does not include or distribute TripoSR model weights.
Users must obtain their own copy of the TripoSR model weights and comply with
the applicable license. After downloading the weights, set their directory
through the MODEL_PATH environment variable as shown above.
The model directory must contain weights.json and all binary tensor files
referenced by that manifest:
triposr_weights/
├── weights.json
├── image_tokenizer_model_embeddings_cls_token.bin
├── image_tokenizer_model_embeddings_position_embeddings.bin
└── ...
weights.json records each tensor name, binary filename, shape, and data type.
FunGT Mesher reads this manifest to locate and load the model tensors.
FunGT Mesher is an independent C++ and SYCL implementation of the TripoSR inference pipeline. It is not affiliated with or endorsed by Stability AI or Tripo AI.